zy839971925-zyy/Agent--deep-research-Workflow-Skills / skills/reasoning-workflow
reasoning-workflow
VerifiedDefault silent workflow for everyday questions, research, writing, design and engineering. Match effort to the request, resolve costly ambiguity, find specialist resources and verify the requested outcome. Explicitly selectable with “调用我的 workflow skills”.
references/semantic-state-contract.md
View on GitHub# Semantic state contract
**Trigger:** Structured or durable state is materialized, machine validation is requested, closure/readiness is being computed, or state from multiple agents/artifacts must be reconciled.
**Reads:** typed canonical state, edge policy, materiality, temporal scope, declared record states, event/raw ledgers.
**Updates:** effective record states, explainable findings, closure/readiness reports, stale/recompute provenance.
**May invalidate:** epistemic closure, delivery readiness, current artifact status, downstream reasoning, verification, effectiveness.
**Exit:** declared state has been interpreted under one shared semantic contract and all blocking findings are explicit.
**Related:** [state model and invariants](state-model-and-invariants.md), [synchronization and recovery](synchronization-and-recovery.md), [traceability and integrity](traceability-and-integrity.md).
**Return:** [root workflow](../SKILL.md) → Durable state / closure.
## Governing principle
**Declared state is input. Effective state is computed.**
A record may declare itself `current`, `answered`, `passed`, or `closed`. That declaration is not authoritative if upstream dependencies, supersession, invalidation, temporal scope, or closure rules imply otherwise.
Likewise:
- `declared_status` is stored on records;
- `effective_status` is computed by the semantic validator;
- `declared_delivery_ready` is a manifest claim;
- `computed_delivery_ready` is validator output;
- work may declare `closed`, but Question-lane closure uses `computed_epistemic_closed`.
Validators certify **structural admissibility, not epistemic truth**. They can prove that records are typed, references resolve, dependency semantics are internally coherent, stale state is not silently reused, and closure rules are satisfied. They cannot prove that a real-world premise is true or that a recommendation is wise.
## Materiality
Every epistemically or operationally relevant record can carry:
`material | supporting | incidental`
The question is not “is this detail interesting?” but:
> If this record were wrong, unknown, stale, or omitted, could it materially change the core answer, judgment, action, safety, or required delivery?
Only material unresolved/stale items normally block closure. Supporting/incidental items can remain bounded with a finding when they do not change the user's outcome.
## Temporal semantics
Do not collapse temporal meaning into one `stale` boolean.
- **invalid:** the record itself is unusable under the contract.
- **stale:** the record may once have been valid but is no longer current for this task/context.
- **superseded:** a newer authoritative version replaces it for the relevant scope.
- **historical:** it remains valid for a past time/scope even if not current for the present task.
Useful fields include `valid_for`, `observed_at`, `effective_at`, and `current_for_task`. Historical validity and present applicability are distinct.
## Record lifecycle families
Different record types have different declared status spaces. Do not reuse one ambiguous universal status enum.
- **Question:** `open | answered | bounded | deferred | invalidated`
- **Answer:** `draft | current | bounded | stale | invalidated | superseded`
- **Premise/Assumption:** `current | challenged | stale | invalidated | superseded | historical`
- **Hypothesis:** `candidate | leading | weakened | rejected | superseded`
- **Uncertainty:** `open | resolved | bounded | irreducible | deferred | invalidated`
- **Inference/Judgment/Recommendation/Decision:** `draft | current | challenged | stale | invalidated | superseded`
- **Artifact:** `draft | current | stale | superseded | historical`
- **Verification:** `pending | passed | failed | stale | blocked`
- **Effectiveness:** `pending | effective | ineffective | inconclusive | stale`
The canonical JSON Schema is the local-shape authority. Cross-record effective-state semantics are implemented by the shared semantic engine.
## Typed reference integrity
Build a registry:
`record_id → {record_type, collection, record, declared_status, effective_status}`
References are valid only when all relevant conditions hold:
1. the target exists;
2. the target has a compatible record type;
3. the target's effective state is compatible with the use;
4. the reference direction matches the edge contract.
For example, `premise_refs` may resolve only to Premise/Assumption records; `question_refs` only to Questions; `verification_refs` only to Verification records.
## Edge policy and cycle semantics
Edge semantics live in one machine-readable policy file (`schemas/edge-policy.json`), not scattered Python conditionals. The policy defines target types, downstream effect, propagation behavior, and cycle policy.
Typical families:
- logical dependency / lineage / supersession: cycles normally invalid;
- causal/feedback relationships: cycles allowed;
- support networks: cycles allowed structurally but may warrant circular-support warnings;
- explicit invalidation: invalidation cycles are normally errors.
`hard` propagation means downstream current-like records become effectively stale/invalid. `recompute` means downstream support/conflict/measurement should be recomputed before material closure. `none/contextual` means no automatic stale propagation; another semantic rule decides.
## Explainable enforcement
Every computed blocking result must explain why. Prefer chains such as:
`P-2 invalidated → INF-3 effectively stale → JDG-4 effectively stale`
Closure reports should enumerate blockers rather than return only a boolean. Findings use:
- **ERROR:** blocks the relevant closure/readiness gate;
- **WARNING:** requires disposition or review but may not block;
- **INFO:** diagnostic context.
[← Return to root workflow](../SKILL.md)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |